If you really believe this, try to use GPT 5.6 to prove an open problem you know nothing about. You might get lucky, but if you don't, you will soon discover that 5.6 can make "progress" towards a theorem without actually getting anywhere pretty much indefinitely.